EUVIMEDEUVIMED entdecken
Crossref

Swst-Net: Semantic-aware wavelet-driven structure and texture interaction network for image inpainting

Lili Shen, Xinglin Wang, Qi Li, Ran Chen, Zhiyao Long

Multimedia Systems · 2026-10 · 32 (9)

DOI: 10.1007/s00530-026-02655-6 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Multimedia Systems, online erschienen am 17.09.2026. Verfasst von Lili Shen und 4 weiteren Autor:innen. Die Arbeit nennt 75 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

To address the challenge of separately modeling structure and texture under semantic guidance and promoting interaction, we propose a semantic-aware wavelet-driven structure and texture interaction network (SWST-Net). Discrete wavelet transform is employed to decompose the image into low- and high-frequency subbands, facilitating the separation of structure and texture information, which is crucial for maintaining both the integrity of image structures and the realism of textures. A dual-encoder and shared-decoder design is used to independently reconstruct structures and textures, while semantic priors from a pretrained MAE are integrated through a semantic-guided alignment module, ensuring that semantic consistency is preserved during the inpainting process. A bidirectional cross-domain feature fusion module further enhances collaborative guidance and dynamic integration between the structure and texture branches, improving the overall inpainting quality. Experiments on the Paris StreetView, CelebA-HQ, and Places datasets demonstrate that SWST-Net consistently outperforms recent methods, achieving significant improvements in both qualitative and quantitative metrics. The results highlight the method’s effectiveness in preserving both structural integrity and texture details, offering a promising solution for high-quality image inpainting tasks.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Shen, L., Wang, X., Li, Q., Chen, R., & Long, Z. (2026). Swst-Net: Semantic-aware wavelet-driven structure and texture interaction network for image inpainting. Multimedia Systems, 32(9). https://doi.org/10.1007/s00530-026-02655-6

Vancouver

Shen L, Wang X, Li Q, Chen R, Long Z. Swst-Net: Semantic-aware wavelet-driven structure and texture interaction network for image inpainting. Multimedia Systems. 2026;32(9). doi:10.1007/s00530-026-02655-6

BibTeX
@article{Shen2026,
  author = {Shen, Lili and Wang, Xinglin and Li, Qi and Chen, Ran and Long, Zhiyao},
  title = {Swst-Net: Semantic-aware wavelet-driven structure and texture interaction network for image inpainting},
  journal = {Multimedia Systems},
  year = {2026},
  volume = {32},
  number = {9},
  doi = {10.1007/s00530-026-02655-6}
}
Kostenlos starten →